Max in conversation with Luo, Head of Artificial Intelligence Solutions and Technologies at Linde
In this episode of Max and the Supply Chain Heroes Max tries to de-mystify the topic of artificial inteligence in combination with supply chain issues.
Due to Corona safety measures he is doing so by remotely interviewing Dexin Luo, an acknowledged expert in Supply Chain Management and artificial intelligence. Together they analyse ways to apply AI to the supply chain in a hands on way. Ludwig Meister themselves are currently taking first steps in using machine learning for ERP purposes, but Max is eager to learn about next steps towards AI usage.
If you like to know how to explain AI on a cocktail Party, listen to today’s podcast episode. Transcript and links can be found below. And as always: we love to hear from you. Leave your comments, suggestions, criticism at email@example.com.
MAX: Welcome to Max and the Supply Chain Heroes. Your entrepreneurial podcast about challenges and changes in procurement and distribution in the context of digitization. Thoughts, experiences and above all findings by experts in supply chain management. Completely free of any consultancy mission, easy understandable, just plained business prospected. I am Max Meister and I hope you enjoy this episode.
MAX: So my/ today’s guest is Dexin Luo. She is working for Linde as head of Artificial Intelligence Solutions and Technologies. And for me, it is interesting to discuss with her about supply chain topics and especially AI in supply chain. And before we go too much in detail, maybe, Dixon, you can say what is your today’s main job at Linde and tell some sentences about Linde as well.
DEXIN LUO: Sure. Hi, Max. Right. I work for the company Linde PLC. It’s a world-largest industrial gas company. And I lead a global team of machine learning engineering data scientists to develop artificial intelligence solutions for the company. We also collaborate heavily with external partners, startups, universities to be able to understand, what are the emerging technologies and how do we apply that to our business to generate value?
MAX: Okay. And when you say the biggest gas company in the world, do you have some figures? How many people are working there? Because I think everybody around Munich knows the company Linde but has probably no image of the company.
DEXIN LUO: I think, today, we have about 60,000 people working in Linde globally. So we are in 100 countries. And we have about a revenue of 28 billion. That’s the number in 2019. We can repeat that maybe.
MAX: No, no. It’s fine. Thank you. When you say you are now working in AI, and if I understand right, and your career station before was/ the biggest topic was supply chain management, what can you/ what are your biggest or what is your biggest focus for AI applications at Linde in improving the supply chain today?
DEXIN LUO: Sure. Yes, so before I started developing AI solutions about more than three years ago, I was in the operational business with a focus on supply chain management and product management. And in the supply chain area in Linde, it’s pretty simple. We try to increase the supply chain efficiency as much as possible. And we try to increase the supply chain responsiveness and make the supply chain as resilient as possible, so risk mitigation. And one of the most important topic in overall Linde is environmental impact. How do we reduce CO2 emission and increase safety of the supply chain operations in general?
MAX: Okay. So I guess that the resilience of the supply chain is quite a big topic at the moment. So we are recording in the midst of May 2020. How do you approach such an AI project within Linde? Because for my listeners, it’s always interesting to really discuss about the project and the steps of such a project. So for me, it would be interesting to maybe discuss. You have a couple of papers everybody can read. So we will put a description below. But what was maybe, supply chain-wise, the last project, and can you describe how you approached this project?
DEXIN LUO: That’s a good question. So first, on the supply chain resilience, so fortunately, the majority of Linde’s business has a relatively regional supply chain. So it’s/ it is affected, but to a much lesser degree than a company with a global supply chain with/ you know, you have/ if you have to procure equipment from China, for example. In the supply chain area, as I have mentioned, we focus this many different topics. Maybe to give you one example how we increase supply chain efficiency, we do have a large internal fleet, and we also outsource a lot of our delivery tasks. We drive more than 1 billion kilometers a year, which is a lot. So to systematically make this delivery more efficient, we have deliver/ we have developed a driver training, too. It sounds very simple. However, it’s powered by a very sophisticated AI algorithm. And how does it work? We use the big data collected on our onboard computer of our fleet to understand all these different parameters, like hash breaking, acceleration, delivery, frequency, and whether it’s inner city or mountain area to understand the correlation between drivers’ driving behavior and fuel consumption. In other words, what can the driver influence on the fuel consumption, and what are the factors that he or she cannot control, right? And we’re able to very systematically provide a tool to be able to train the driver real time how he can drive better to reduce the fuel efficiency with a gamification element as well.
MAX: But when you say you want to increase the drivers’ efficiency, can you give an example what this means? This means just the consumption of gas or also time, or what does it mean?
DEXIN LUO: It’s mainly about gas. So let’s say, today, I’m sure in many logistics company, you do do driver training, and you have some kind of scoring. But it’s very hard to touch the overall driver community. Often, when the transport manager is trying to tell the driver, “Hey, you should have braked less or did less acceleration,” it’s very conceptual, right? The driver might say, “That’s not fair because, on my trip, there were a lot of congestion, or I was driving in a mountain area. It’s very different from my colleague, who had a very different setting,” right? But with this tool, we’re able to make a very fair evaluation to predict with high accuracy because we’re able to single out the elements which the driver cannot control. So you’re able to, let’s say, evaluate and compare the driver on the very fair basis. Now I think this is very important because, to me, the driver community is a very important community within the supply chain management. And when we talk about AI, there are many misunderstanding. And here, we aim to help them, right? We/ and for them, many of them might not want to understand a lot of complex statistics. And in reality, we actually want to simplify the process, right? So instead of going through many different numbers with them, which they don’t have time because they are usually in a very tight schedule, we show that our tool actually can provide a very fair and accurate prediction of how their driving style influence the fuel consumption, right? So it really is over the long time. Over a longer time period, you build this trust on the driver, on the tool. And so the driver believes in the evaluation. And additionally, this gamification aspect helps the driver to incentivize him to actually drive better.
MAX: Okay. So if I understand right, you analyze with an algorithm all the data you have, and you try to find some improvements in driving, and then you have a special simulator where the driver is tested, or do you do this in real time in the gas trucks?
DEXIN LUO: We do that after each trip.
MAX: Ah, okay. So you/ when a trip is completed, you analyze it, and then you give feedback to the driver.
DEXIN LUO: Yeah. But the driver can see it on his own device. So we don’t really have to manually do this education.
MAX: Okay. Okay. So I think this is an interesting example. What I think is always difficult when you talk about artificial intelligence is that, for a nontechnician like I am, it’s a black box. So I think it’s totally normal that people having no clue about AI that they are/ they have a big respect for this topic, and they want to/ yeah, the trust has to be earned by the system, so not the other way around. So I can imagine this was one of the biggest topics you had in this project?
DEXIN LUO: Yes, I think I wanted to talk about this project is also that it shows that AI is not making things more complicated by rather simplify. So imagine/ I mention a lot about this project, but from a driver perspective, he is only receiving an app which after each trip shows him a score, which is with some kind of gamification, which is very simple to understand. It’s like if you play a computer game, and it just shows you how good you have played.
MAX: Okay. So this example is maybe not the best one for Ludwig Meister because, in Munich, we only have one driver. So I think the gain of efficiency wouldn’t be high enough. But before we discuss another example, I wanted to come back to the topics where you think AI can add value in the supply chain. You said you have/ you want to increase efficiency. This was the drivers example. And then you have the reduction of risks or the improving the resilience. Do you have one topic where Linde is now making some projects on, or do you have a/ some thoughts on this topic?
DEXIN LUO: On the resilience side, of course. So traditionally, we have always focused on safety and resilience. And here, with AI’s help, you can do many things. For example, we have done a project in the areas of safety, transport safety. So we’re able to use a lot of data, not only including Linde’s internal data, but external traffic, you know, road construction, right, the road characteristics and also the general population’s driving history, like whether there was an accident, to predict whether/ when a driver is entering a certain area, what’s the probability of entering into an accident so that you can avoid that, right? So similarly, I know that, for many companies, probably you don’t have so many driver, but similarly, what we have been doing is that you could predict the most important risk factors in your whole supply chain with the help of a lot of external data. So in your example, for example, you could think about, hey, if I am procuring equipment from a certain country, right? And for me, it’s blind what’s happening and what can influence the risk of supply. And here, you can use a lot of external data to be able to predict your supplier risk, to be able to predict the risk of that country or even the particular city where the production is being made, and having a correlation to the overall supply chain risk of your company.
MAX: Yeah, I think this is a good example. For us, at the moment, what we have done is we have/ but it’s manual work. We created some kind of risk factors for the different countries. We are not sourcing that much in China, but we have many suppliers also in Italy, for example. And then we created a/ it’s a factor. And we tried to calculate the supply chain risk. But it’s/ basically, it’s Excel sheet logic. So it’s not using real-time data. It’s manual work. So I think this could be an important topic for the future. So I think you’re definitely right, yeah.
DEXIN LUO: Yeah.
MAX: When we talk about/ so maybe because/ how/ so when you are at a cocktail party and somebody asks you, “What are you doing?” how do you explain AI? And if somebody is asking, “I want to use it for my company,” or, “I have a special idea. How should I start?” so how do you explain it? Because I imagine it’s/ you will be asked this sometime?
DEXIN LUO: I think, like if you’re in a cocktail party, the easiest way to explain what AI is, is face recognition, right, when you’re trying to open your iPhone or Alexa voice assistant, right? That’s what people relate most with this type of technologies. And we do use this type of computer vision technology also to be able to recognize our assets and to make a supply chain more transparent. And that would be examples I could give. I think, right, so perhaps, to go to one area in the supply chain area which everybody can relate to is demand forecast, right? So you know that supply chain has this bullwhip effect, and the demand forecast accuracy is very important. The more accurate you are and the better you can plan, right, in general. This is exactly why AI can help. In the past decades, people have been doing this all along. And why can’t AI do better? It’s mainly, as I mentioned, also maybe because of two most important factor. One is a lot more data available internally with IoT sensor, also externally that you can utilize, whether events/ for example, we live in Munich. The Oktoberfest is a big event which influence a lot on our demand of CO2. And such information you can put into a demand forecasting algorithm. So that’s the/ the algorithm can learn to understand a lot more dynamics that influence your customer demand. So I think this is maybe one example which I would give.
MAX: Okay. So I think Oktoberfest is a good example. So the CO2 is used in restaurants and bars for the tapping, for tapping a beer or stuff like this?
DEXIN LUO: Yeah.
MAX: Okay. When you say you’re using external data, what is more important for the demand planning, for example? Is it the history/ your sales history, or is it the external data? What do you think?
DEXIN LUO: It depends largely on the customer, right, and the product. So you will see some product has a much more seasonality or is much more influenced by external events like the Oktoberfest. And some products are very regularly driven. So therefore, which is exactly where AI can help because you’re able to segment the customer and tailor-make the forecast into each of them, right, and to learn from many different data that you can increase the accuracy very high.
MAX: Okay. So you just mentioned customer segmentation. And I think this is one of the biggest topics for our company because we have very different kinds of customers, so for example, a paper machine company or a production company of pumps and motors. So the demand is very different, or it’s unique for each kind of company. If I want to use an AI algorithm in my company that helps me identifying, for example, similar segments of customers, how would you start such an AI project?
DEXIN LUO: So I think the first question I would always ask you as a business owner would be, why do you want to segment the customers? Is it that you want to make better supply chain planning? Is it that you want to develop different marketing strategy or product service offering to these different customers? I understand it’s more on the supply chain side, right?
MAX: Yes, yes. For us, it’s/ we want to add value within the supply chain for our customers. So we have to have the right services and products at the time where our customer needs the product or the service, yeah.
DEXIN LUO: Yeah. So right. So in this respect, the reason why I ask why is you have to/ at the end of the day, I hope it’s clear for everyone we are mainly talking about software development, right, even though there’s all these fancy algorithms behind. So the reason why I ask why is because you have to understand if/ suppose I can/ forget about how. Suppose I can give you a perfect segmentation. How would you use it? What actions would you take? Who is the end user of such a software, right? This is the most important thing. And you have to know your company DNA and your process, right? So and then you can figure out/ you can/ I suppose if the/ depending on the company size, you can hire a data scientists or find a data science team externally, find some company to help you figure out how to segment. But we can go to the technical details. But you know, it’s actually not so complicated, right? So for example, you might have different SKU. And there, you want to decide, okay, what are the features that influence my supply, how to differentiate them? Could be the size of the SKU, could be, you know, different delivery frequency, could be different customer buy-in belief, etc., etc. And then you/ yeah, the rest/ so AI, people say that it’s largely about feature engineering. That is actually domain expertise is needed from your business. And the rest is not so complicated. So there are many different clustering algorithms, like K means, K prototype, etc., etc. And then afterwards, once you build the segmentation, you have to try to interpret what that means. And this is an iterative innovation process. So in summary, I would say, how do you do AI? You first of all try to understand why you do that. And you can use a process like design thinking to be able to figure out, what is my end solution and end user who can act on my software? And then you do an iterative innovation process to improve it.
MAX: Okay. Okay. So I understand, and I think this is maybe the biggest learning of the last 20 minutes. If you want to use AI in your company, it’s not about discussing about it or thinking about it what we could do, but it’s very important to really ask why you want to do it and ask the question what kind of service you want to develop or what kind of value you want to add. And after you have done this, and this is really I think the biggest question, then you should start searching for the right people, or you maybe have it in house because a data scientist always sounds very big, but if you really start looking at your own data, I’m quite sure that you already can see a lot of things. But you have to invest the resources and the time to really look at the data. And just after this step, you would try to really use AI algorithms, or I have no clue how they work, but then you would try to use it. And then the interpretation of the data in the end really adds the value and is the last step. Is it right?
DEXIN LUO: Yes, for sure. I think, in general, in this time, technology is developing extremely fast. It’s always good to have some internal know-how to understand where to look for what and what would be a good partner. The rest is exactly as you said.
MAX: Okay. So maybe one last question. If you have somebody who is/ going back to the cocktail party, do you have a/ some kind of Website or some content where you really can refer to where somebody’s able to learn more or get some more details, or how do you/ how are you inspired?
DEXIN LUO: That’s a good question. So I do a lot of different things. You know, for anybody, the simplest thing is you can set up a Google alert to know, what are the latest AI news? And there are many different communities, Websites. Like for example, Towards Data Science is a good one. It’s a good blog where a lot of the top data scientists are posting. And there are a lot of open source free course you can take. And in many cities, there are meetups. So in Munich, for example, there’s a community. There are meetups you can go to to exchange with peers and to understand more about the technology. So yes, there are many things you can do.
MAX: Okay. So I would be happy if we can put the description of the blog below. And whenever this corona crisis is gone, maybe we can meet during a meetup here in Munich. So I would be happy to learn more.
DEXIN LUO: That sounds great. Hopefully also during the Oktoberfest time.
MAX: Yeah, then maybe next year. So we will see. But yeah, Dexin, so thank you very much for your insights. And whenever one of the listeners has some more questions, so I would give the contact, or everybody can send me an email to firstname.lastname@example.org. And I will maybe pass on the questions. So we will make a second show. So I’m very grateful. Thank you for your time and all the insights.
DEXIN LUO: Thank you very much, Max, for inviting me.